Fault Diagnosis For Wind Turbine Gearboxes Based On The Cooperative Convolutional Neural Networks

With the rapid development of my country's wind power industry,a large number of wind turbines installed during the rapid development period have gradually come out of the warranty period,and wind power operation and maintenance work has increasingly become the focus of industry development.However,wind turbines are widely distributed and remotely located,the traditional operation and maintenance mode has high operation and maintenance costs and low operation and maintenance efficiency.At present,in order to more effectively monitor the health status of wind turbines,the monitoring sensors arranged in wind turbines are becoming more intensive and the amount of monitoring types is becoming more and more abundant.The scale of wind power monitoring data is showing a blow-up growth.It provides strong data support for the healthy service status of wind turbines,improving operation and maintenance efficiency,and reducing operation and maintenance costs.However,the big data of wind turbines has the characteristics of various types of monitoring quantity,large span,data redundancy and low value density,which make the operation and maintenance of wind power more difficult and demanding.As a result,the traditional operation and maintenance methods based on signal processing and traditional machine learning encountered low efficiency and accuracy,which has been unable to effectively mine the healthy information of wind turbines in the big data.As a potential intelligent analysis and processing method of big data,deep learning can learn the effective features of massive data end-to-end adaptively by building a deep model,mining the rich potential information in the data,and providing a new idea for intelligent operation and maintenance of wind turbines.As one of the key components of wind turbines,wind turbine gearboxes have a relatively low frequency of failures,but the maintenance time for each failure is long,which seriously affects the economic benefits of wind power generation.Convolutional neural network(CNN)is a very representative deep learning method,which can adaptively mine the discriminative fault features of wind turbine gearboxes from high-dimensional data,and characterize the complex and changeable state information hidden inside the data.As the wind tubrbine gearboxes have the characteristics of weak fault characteristics,multi-sensor monitoring,and few labeled fault samples,it is difficult to directly use a single CNN to meet the needs of wind turbine gearbox fault diagnosis under different characteristics.Therefore,aiming at the problem of fault diagnosis of wind turbine gearboxes,this paper made full use of the adaptive features extraction ability of the CNN,and proposed the collaborative convolutional neural network fault diagnosis method,which includes the field enhanced convolutional neural network(RFECNN),the dynamic integrated convolutional neural network(DECNN)and the semi-supervised convolutional neural network(SSCNN).The main research work of this paper is as follows:(1)Due to the complex structure,changeable working condition and the disturbance of strong background noise,the fault features of wind turbine gearboxes is weak and difficult to extract.Aim at this problem,a fault diagnosis method called the RFECNN was proposed.This proposed method directly takes the original vibration data as the input of the model,and through adding the one-dimensional convolutional layer and the dilated convolution operation on the traditional two-dimension convolutional neural network to enhance the receptive field of the model,which makes it easier for the model to capture the fault information in the long-term dependence of vibration signal,enhances the fault feature extraction ability of the model,and improves the accuracy of fault diagnosis of the wind power gearbox accuracy rate.(2)Due to the vibration signals captured by a single sensor are often difficult to fully and accurately represent the health status of the wind power gearbox.Aim at this problem,a multi-sensor fusion fault diagnosis method of the wind turbine gearbox based on the DECNN was proposed.This method combines the advantages of the ensemble learning and the convolutional neural network,which not only considers the problem that different vibration sensors contain inconsistent fault information due to the influence of installation location,transmission path and fault location,but also fully utilizes the advantages of strong feature learning ability of multiple sub convolutional neural networks.As a result,the proposed method can effectively integrate the information collected by different vibration sensors to further improve the fault diagnosis accuracy of wind turbine gearbox.(3)As it is difficult to get enough labeled fault samples for some wind fields,the problem of over fitting and low classification accuracy is easy to occur when using convolutional neural network for fault diagnosis.Aim at this problem,the SSCNN was proposed for fault diagnosis of wind power gearbox.This method has not only inherited the strong feature extraction ability adaptive of the traditional convolutional neural network,but also introduces the domain adaptation method to adapt the distribution of labeled samples and unlabeled samples.As a result,the proposed method can simultaneously use the label information of labeled samples and the fault information of unlabeled samples from the feature distribution space,which can help promote the semi-supervised convolutional neural network to extract more discriminative fault features,reduce the dependence of traditional convolutional neural network on expensive data labels,and improve the generalization ability of the model.(4)According to different application requirements,the proposed collaborative convolutional neural network fault diagnosis method was not only integrated on the network monitoring and fault diagnosis system,but also deployed in the in big data platform of wind turbines.Then,the different system modules are verified by examples.Finally,at the end of this paper,the previous research work is summarized and the future research work is prospected.